基于LSTSVR模型的边缘计算预测变压器平均油温及绕组热点温度

变压器绕组的热点温度过高,会导致变压器绝缘脆解、裂化甚至击穿短路。因此及时、准确地预测出变压器绕组的热点温度,对提高变压器运行的安全可靠性至关重要。利用最小二乘双支持向量回归机(LSTSVR)作为边缘计算模型,将变压器油中气体色谱分析数据信息与变压器负载电流、环境温度、顶层油温、上死角温度等变压器运行信息结合,构建监测系统架构,预测变压器的平均油温,并计算出绕组热点温度。将所提方法得到的数据与实测数据进行对比,结果利用LSTSVR模型实现了变压器平均油温及绕组热点温度的准确预测,且该模型的预测精度优于最小二乘支持向量回归机模型,有效地提高了绕组热点温度测量的精度。现场实例也证明了所提方法的有效性和可靠性。

Prediction of transformer average oil temperature and winding hot spot temperature by edge computation based on LSTSVR model

The excessively high hot spot temperature of transformer windings leads to transformer insulation embrittlement and cracking, even breakdown and short circuit. Therefore, it is very important to predict the hot spot temperature of transformer windings timely and accurately to improve the safety and reliability of the transformer operation. Based on the edge computation model of LSTSVR(Least Square Twin Support Vector Regression) and the DGA(Dissolved Gas Analysis) information combined with transformer operation information such as transformer load current, ambient temperature, top layer oil temperature and upper dead angle temperature, a monitoring system is constructed to predict the average oil temperature of transformer and calculate the hot spot temperature of windings. The comparison between data gained by the proposed method and the field measured data shows that the LSTSVR model realizes the accurate prediction of the transformer average oil temperature and hot spot temperature of the windings, and the prediction accuracy of the LSTSVR model is better than that of the LSSVR(Least Square Support Vector Regression) model, the accuracy of hot spot temperature measurement of transformer windings is improved effectively. The effectiveness and reliability of the proposed method are proved by field examples.